New Sub-Nanosecond Timing for In-Beam PET in hadrontherapy · 2012. 1. 21. · LOR 2 coincidence 1 2 LOR 1 2 + annihilation:two 511 keV photonsemitted back to back ˇ180 . Coincidence

Sub-Nanosecond Timing for In-Beam PET in hadrontherapy

Sub-Nanosecond Timing for In-Beam PET inhadrontherapy

Baptiste Joly

Clermont Université, Université Blaise Pascal, CNRS/IN2P3, Laboratoire de PhysiqueCorpusculaire, BP 10448, F-63000 CLERMONT-FERRAND, France

University of Chicago, July 15, 2010

1 / 33


Outline

1 In-beam Positron Emission Tomography for treatment verification inhadrontherapy

HadrontherapyReal-time monitoring of ion ballisticInterest of Time-Of-Flight PET

2 Technological factors determinig time resolutionDetection processExperimental set-upComparison of scintillatorsComparison of timing algorithms

2 / 33


Outline

Introduction

• Hadrontherapy is raising interest for the treatment of certain tumors.

• Need for treatment verification systems.

• Positron Emission Tomography is a promising technique for thisapplication.

• Instrumentation development is required to adapt the technique.

• Time Of Flight (TOF): a key point for performance, and a technologicalchallenge.

3 / 33


In-beam Positron Emission Tomography for treatment verification in hadrontherapy

1. In-beam PET for treatment verification inhadrontherapy

4 / 33



Hadrontherapy

A technique for inoperable and radioresistant tumors

Localized (58%)

surgery only 22%radiotherapy only 12%surgery and radiotherapy 6%inoperable and radioresistant 18%

Metastatic (42%) chemotherapy 5%palliative treatment 37%

• Cancer: 2nd cause of death in the West.• ≈ 18% of localized tumors are both:

• Inoperable, close to organs at risk.• Radioresistant for conventional radiotherapy.

• Hadrontherapy is suited for those tumors because of the properties ofion-matter interaction.

5 / 33



Hadrontherapy

Ionisation properties and biological effect

• Dose distribution:• Photons, electrons: dose decreases

with depth.• Ions: maximum at Bragg peak.

• Dense ionisation in the trajectory⇒high biological efficiency.

• During a treatment, the energy ismodulated⇒ Spread-Out BraggPeak (SOBP).

• Effective dose profile for several ions:• Dose (SOBP) > dose (entrance

plateau).• Tail: radioactive fragments.• Carbon: adapted to hadrontherapy.

6 / 33



Hadrontherapy

Operation

• Passive shaping:• Lateral scattering.• Energy dispersion.• Compensator: modulates

energy.

• Active shaping:• Magnetic deviation: lateral

scanning (x − y ).• Energy modulation: depth

scanning layer by layer.

7 / 33



Hadrontherapy

Nuclear fragmentation

• Collisions ions - nuclei of the bio. medium⇒ fragmentation (≈ 50% of Cions at 300 MeV/u).

• ⇒ Prompt and slow activity.• Abrasion-ablation model:

• Collision with impact parameter b.• Abrasion: formation of a “fireball”, target and projectile fragments.• Ablation (ou evaporation): de-excitation, emission of n, p, γ.• Radioactive nuclei produced.

• Dispersion of dose after Bragg peak.

• Possibility to detect γ or β+ activity⇒ PET.

8 / 33



Real-time monitoring of ion ballistic

Detecting β+ activity to control the ion ballistic

P. Crespo 2005, PosGen simulation.

• Fragmentation⇒ β+ nuclei,• Projectile fragments: activity

concentrated at the end ofthe traject.

• Target fragments: spreadthe activity.

• 11C predominant(T=20 min).

radionuclide half-life11C 20.4 min15O 2 min12N 11 ms10C 19.3 s8B 770 ms

• Activity correlated with dose,maximal at Bragg peak.

• ⇒ In-beam PET.

9 / 33




PET principle

coincidence

• β+ annihilation: two 511 keV γphotons emitted back to back≈ 180◦.

• Coincidence detection (if|t1 − t2| < time window).

• Recording of a line ofresponse (LOR).

• Parasitic events:• Scattered pairs (30-40% of

annihilation pairs).• Random pairs, high rate for

in-beam PET (nuclear γ).

10 / 33




PET principle

e+ + e− → 2γcoincidence







10 / 33




PET principle

e+ + e− → 2γ

γ1

γ2

coincidence







10 / 33




PET principle

e+ + e− → 2γ

γ1

LOR

γ2

coincidence







10 / 33




PET principle

e+ + e− → 2γ

γ1

LOR

γ2

coincidence

γ1

γ2







10 / 33




PET principle

e+ + e− → 2γ

γ1

LOR

γ2

coincidence

γ1

γ2

LOR







10 / 33




PET principle

e+ + e− → 2γ

γ1

LOR

γ2

coincidence

γ1

γ2

LOR

γ1

γ2







10 / 33




PET principle

e+ + e− → 2γ

γ1

LOR

γ2

coincidence

γ1

γ2

LOR

γ1

γ2

LOR







10 / 33




Experience on in-beam PET at GSI, Darmstadt

Example: BASTEI (GSI)

• Two blocks from a commercial camera(ECAT EXACT, CTI).

• System modified to stamp the events:• Beam on (1500 cps)⇒ noise.• Beam off (200 cps)⇒ reconstruction.

• Verification after the irradiation.

Necessary developments

• Geometry (sensitivity, artefacts).

• Rejection of randoms, beam on.

• “Real-time” verification (<session).

11 / 33




Treatment verification process at GSI

12 / 33




In-beam PET: a challenge

• Limits of BASTEI-like systems• Low β+ activity

• Clinical PET, radiotracer: 10-100 kBq cm−3.• In-beam PET: 200 Bq Gy−1 cm−3⇒ a few kBq cm−3.

• β+ activity is rapidly “washed out” by metabolism (≈4 min)⇒ “in-beam”acquisition necessary.

• In hadrontherapy, the nb. of irradiation fractions tends to 1⇒ verificationmust be done during one fraction.

• Hight parasitic activity (γ, neutrons, p, e−).• The new beams are continuous, i.e. without “macro” pause⇒ the

acquisition must be synchronized with beam at ns time scale to rejectparasitic prompt particles (≈ 1 ns after fragmentation).

• Benefits of Time-of-Flight:• Better exploitation of the low statistics,• Better rejection of parasitic particles.

13 / 33



Interest of Time-Of-Flight PET

Time of Flight: principle and benefit

β+

γ1, t1

γ2, t2

• t1 − t2 ⇒ localization along the LOR• Time resolution ∆t ,• Localization ∆x = c/2 ∆t ,• Example 500 ps −→ 7.5 cm.

• Better rejection of randoms.• Better image quality by reducing the

coupling btw. voxels:• Smaller statistical noise (factor

D/∆x),• Example: whole body PET,• ∆x = 7.5 cm, D = 40 cm,• ⇒ Improvement factor F = 5.

• Reconstruction: faster convergence.

• Time of flight is the industrial state ofthe art of recent clinical PET systems.

14 / 33


Technological factors determinig time resolution

2. Technological factors determinig timeresolution

15 / 33



Detection process

Detection process

16 / 33



Detection process

Inorganic scintillators for PET

Scintillation mechanism

• Photoelectric or Compton interaction.

• Secondary ionisations in cascade.

• Excitation of luminescent centres.

• Radiative de-excitation 400-500 nm, decaytime=some 10 ns.

• Random emission times⇒ statistical limit totime resolution.

Candidate materialsname attenuation length PE light decay

at fraction yield time511 keV (mm) (%) (ph/keV) (ns)

LSO 11.4 32 30 40LYSO 12 32 41LPS 14.1 29 20 30LuAP 10.5 30 11 18(90%)LaBr3 (h) 22.3 13.1 70 16LaCl3 (h) 28.0 14.7 46 25(65%)LuI3 (h) 18.2 28 95 24(60%)

drawbacks (h):

hygroscopic

advantages

17 / 33



Detection process

Photodetectors: today

Photomultiplier tubes(PMT)

• Only photodetectors used inclinical PET until now.

• Advantages: fast, high gain.

• Drawbacks: dimensions⇒block detector with position“decoding”.

Detector block

• Light sharing btw. 4 PMT,

• Position reconstructed from charge ratios,

• Light loss and propagation path limit time resolution.

18 / 33



Detection process

Compact photodetectors

Micro-Channel Plate PhotoMultiplier Tubes (MCPPMT)

+ High gain (105-106),+ Very fast response,- Cost of commercially available models,- Aging.

Avalanche Photo-Diode(APD)+ High quantum efficiency (70-80%),+ Low cost,- Noise,- Low gain (50-200).

Geiger-mode APDmatrices (SiPM)

+ High gain (105-106),+ Fast response,- Noise,- Stability T◦ and Vpol .

19 / 33



Detection process

Signal read-out

20 / 33



Detection process

Digital front-end concept

Avantages compared to analog circuits

• Generic scheme,

• Reconfigurable,

• Versatile,

• Stability: baseline shift correction,

• Piled-up events can be handled.

21 / 33



Experimental set-up

Two detectors in coïncidence

• Channel 1: “fast”, reference channel,LaBr3 (16 ns, 63 ph/keV).

• Channel 2: “test channel”, here LYSO(41 ns, 32 ph/keV).

0 20 40 60 80 100

−4

−2

0

ch1

ch2

time (ns)

volta

ge(V

)

• Fast PMTs (rise ≈ 700 ps).

• Oscilloscope Bandwidth=4 GHz,Sampling Rate=10 GSps.

• Algorithm⇒ event energy and time.

22 / 33



Experimental set-up

Data Processing

2.5 3 3.5 4 4.5 5 5.5·105

R = 3.3%

selected batch

LaBr3

charge (a.u.)

• Event selection on energy (±2.5σ).

• First measurement: LaBr3 on bothchannels, fwhm1−1 = 237 ps.

• Second measurement: LaBr3 on ch1,LYSO on ch2, fit gives fwhm1−2.

• Meaningful figure: coincidenceresolution for 2 detectors like ch2fwhm2−2 =√

2× fwhm1−2 − fwhm1−1.

23 / 33



Experimental set-up

Data Processing

1 2 3 4 5 6·105

R = 12.7%

selected batch

LYSO

charge (a.u.)

• Event selection on energy (±2.5σ).




2× fwhm1−2 − fwhm1−1.

23 / 33



Experimental set-up

Data Processing

−0.2 0 0.2 0.4 0.6t1 − t2 (ns)

dLED, 4789 evts

Gaussian fit, σ = 101 ps• Event selection on energy (±2.5σ).




2× fwhm1−2 − fwhm1−1.

23 / 33



Comparison of scintillators

Crystal shape and reflector

• Test channel 2: LYSO crystal of different shapes and surface state.• In each case, we measure:

• Time resolution,• Peak of amplitude distribution ∝ nb of photoelectrons n,• Light yield is normalised by the best configuration, n0.

• Time resolution is normalised by√

n0/n.

dimensions reflector relative nb. of t-resolution fwhm2−2 (ps)length coupled phe− measured normalized(mm) area (mm2) n/n0 ×

√n/n0

4 4×22 white painting 1 339 3394 4×22 none 0.82 384 3484 4×22 black paint. 0.22 626 292

22 4×4 white paint. 0.43 461 30422 4×4 none 0.56 436 32822 4×4 Teflon tape 0.77 359 31522 4×4 aluminum sheet 0.39 450 28322 5×5 Teflon 0.83 368 3362 2 × 10 white paint. 0.93 299 288

10 10 × 10 white paint. 0.99 350 348

24 / 33




Correlation between light yield and time resolution

0.2 0.4 0.6 0.8 1

300

400

500

600

700

relative light yield(n/n0)

t-res

olut

ion

fwhm

2−2

(ps)

t-resolution vs relative light yield

mesures318 ×

√n0/n • Relation in 1/

√n confirmed.

• No extra effect of lightpropagation time in longcrystals.

25 / 33




Comparison of LaBr3 crystals with increasing ceriumconcentration

% Ce relative nb. t-res. fwhm2−2 (ps)of phe− measured normalized ×

√n/n0

5 1 255 25510 1.11 236 24920 1.30 160 18230 0.62 194 152

−2 0 2 4 6 8

0

0.2

0.4

0.6

0.8

1

time (ns)

mea

npu

lse

(a.u

.)

5%10%20%30%

• Rise time decreases withincreasing Ce concentration.

• Light yield changes must becorrected for.

• Normalized t-resolution isimproved.

• Problem: high Ceconcentration makes thecrystal brittle.

26 / 33



Comparison of timing algorithms

Timing algorithms

Leading Edge Discriminator (LED)

4 6 80

t

time (ns)

volta

ge(a

.u.)

• Search the time when signal crossesthreshold.

• Fine time by interpolation.

• Sensitive to amplitude fluctuation.

Constant Fraction Discriminator (CFD)

5 10 15

0

time (ns)

volta

ge(a

.u.)

attenuatedinverted, delayed

sum

• Search the time when bipolar signalcrosses ground level.

• Insensitive to amplitude fluctuation.

27 / 33




Results

0 0.1 0.2 0.3 0.4 0.50

0.2

0.4

0.6

0.8

1

1.2

relative dCFD threshold

fwhm

2−2

(ns)

dLEDdCFD

0 0.2 0.4 0.6 0.8 1dLED threshold (mV) • Results very similar with dLED

/ dCFD.

• Cause: amplitude fluctuation� shape fluctuation.

• Optimal threshold ≈ 6-8%.

• Time reconstructed by leastsquares fit of the pulse with areference shape:fwhm2−2 = 552 ps.

• The time information is carriedby the initial part of the risingedge (first photoelectrons).

28 / 33




Effect of low-pass filtering

5 10 15 20

0

time (ns)

sign

al(a

.u.)

raw signalfiltered, c = 4filtered, c = 8

filtered, c = 16filtered, c = 32

0 0.1 0.2 0.3 0.4 0.50

0.2

0.4

0.6

0.8

1

relative thresholdfw

hm2−

2(n

s)

dCFDc = 2c = 4c = 8c = 16c = 32

• Optimal low-pass filtering c ≈ 5: little improvement.

• Results degrade if frequency cut (3dB) < 1GHz.

29 / 33




Effect of sampling rate and ADC resolution

Sampling rate

109 10100

0.2

0.4

0.6

sampling rate (Hz)

fwhm

2−2

(ns)

t-resolution vs sampling rate

Linear interp.Cubic spline interp.

• Signal is downsampled at freq. F/n.

• Strong dependence at F < 1.5 GSps.

• Little improvement beyond.

• Curve interpolation useful whenF ≈ 1 GSps.

ADC Resolution

4 5 6 7 80

0.2

0.4

0.6

nb of bitsfw

hm2−

2(n

s)

t-resolution vs nb. of bits

F=10 GHz, dCFDF=1 GHz, dCFD, spline int.

• 5 bits are enough.

• 4 bits at F = 10 GSps.

30 / 33




Conclusions

• In-beam TOF PET⇒ instrumentation challenge.• Time resolution is limited fundamentally by the scintillation process:

• Light yield and time constants are crucial.• The information is carried by the first photoelectrons.• T-resol. ∝ 1/

√n (nb. of phe−)⇒ a gain is possible on light collection

efficiency and photodetector quantum efficiency.

• MCPPMT development is promising for PET: large area, fine positionreconstruction, high gain and fast response.

• The recent developments in fast sampling electronics make possible aTOF PET system with digital signal readout.

• Simple and performant algorithm proposed: low-pass filter and constantfraction discriminator, with ajusted parameters.

31 / 33




Perspectives

• In-beam measurements at GANIL ion cyclotron, Caen, France (firstexperiment done, analysis soon):• Count rates ?• β+ emitter production rate ?• Possibility to discriminate β+ and prompt γ events ?• Specifications for a dedicated electronics ?

• Collaborations involving Clermont-Ferrand:• National scale: GdR MI2B / WP9 Contrôle de dose en ligne (in-beam dose

monitoring).• 7th European Framework Prog. / ENVISION European NoVel Imaging

Systems for ION therapy.• Large Area PhotoDetector (LAPD) project, use of Micro-Channel Plate

PMTs.

32 / 33




Thank you for attention

33 / 33

Documents

New Sub-Nanosecond Timing for In-Beam PET in hadrontherapy · 2012. 1. 21. · LOR 2 coincidence 1 2 LOR 1 2 + annihilation:two 511 keV photonsemitted back to back ˇ180 . Coincidence